kernel/fork.c

   1 /*
   2  *  linux/kernel/fork.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  *  'fork.c' contains the help-routines for the 'fork' system call
   9  * (see also entry.S and others).
  10  * Fork is rather simple, once you get the hang of it, but the memory
  11  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
  12  */
  13
  14 #include <linux/slab.h>
  15 #include <linux/init.h>
  16 #include <linux/unistd.h>
  17 #include <linux/module.h>
  18 #include <linux/vmalloc.h>
  19 #include <linux/completion.h>
  20 #include <linux/personality.h>
  21 #include <linux/mempolicy.h>
  22 #include <linux/sem.h>
  23 #include <linux/file.h>
  24 #include <linux/fdtable.h>
  25 #include <linux/iocontext.h>
  26 #include <linux/key.h>
  27 #include <linux/binfmts.h>
  28 #include <linux/mman.h>
  29 #include <linux/mmu_notifier.h>
  30 #include <linux/fs.h>
  31 #include <linux/nsproxy.h>
  32 #include <linux/capability.h>
  33 #include <linux/cpu.h>
  34 #include <linux/cgroup.h>
  35 #include <linux/security.h>
  36 #include <linux/hugetlb.h>
  37 #include <linux/swap.h>
  38 #include <linux/syscalls.h>
  39 #include <linux/jiffies.h>
  40 #include <linux/futex.h>
  41 #include <linux/compat.h>
  42 #include <linux/kthread.h>
  43 #include <linux/task_io_accounting_ops.h>
  44 #include <linux/rcupdate.h>
  45 #include <linux/ptrace.h>
  46 #include <linux/mount.h>
  47 #include <linux/audit.h>
  48 #include <linux/memcontrol.h>
  49 #include <linux/ftrace.h>
  50 #include <linux/proc_fs.h>
  51 #include <linux/profile.h>
  52 #include <linux/rmap.h>
  53 #include <linux/ksm.h>
  54 #include <linux/acct.h>
  55 #include <linux/tsacct_kern.h>
  56 #include <linux/cn_proc.h>
  57 #include <linux/freezer.h>
  58 #include <linux/delayacct.h>
  59 #include <linux/taskstats_kern.h>
  60 #include <linux/random.h>
  61 #include <linux/tty.h>
  62 #include <linux/blkdev.h>
  63 #include <linux/fs_struct.h>
  64 #include <linux/magic.h>
  65 #include <linux/perf_event.h>
  66 #include <linux/posix-timers.h>
  67 #include <linux/user-return-notifier.h>
  68 #include <linux/oom.h>
  69 #include <linux/khugepaged.h>
  70 #include <linux/signalfd.h>
  71
  72 #include <asm/pgtable.h>
  73 #include <asm/pgalloc.h>
  74 #include <asm/uaccess.h>
  75 #include <asm/mmu_context.h>
  76 #include <asm/cacheflush.h>
  77 #include <asm/tlbflush.h>
  78
  79 #include <trace/events/sched.h>
  80
  81 #define CREATE_TRACE_POINTS
  82 #include <trace/events/task.h>
  83
  84 /*
  85  * Protected counters by write_lock_irq(&tasklist_lock)
  86  */
  87 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  88 int nr_threads;                 /* The idle threads do not count.. */
  89
  90 int max_threads;                /* tunable limit on nr_threads */
  91
  92 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  93
  94 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
  95
  96 #ifdef CONFIG_PROVE_RCU
  97 int lockdep_tasklist_lock_is_held(void)
  98 {
  99         return lockdep_is_held(&tasklist_lock);
 100 }
 101 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 102 #endif /* #ifdef CONFIG_PROVE_RCU */
 103
 104 int nr_processes(void)
 105 {
 106         int cpu;
 107         int total = 0;
 108
 109         for_each_possible_cpu(cpu)
 110                 total += per_cpu(process_counts, cpu);
 111
 112         return total;
 113 }
 114
 115 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 116 # define alloc_task_struct_node(node)           \
 117                 kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node)
 118 # define free_task_struct(tsk)                  \
 119                 kmem_cache_free(task_struct_cachep, (tsk))
 120 static struct kmem_cache *task_struct_cachep;
 121 #endif
 122
 123 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
 124 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 125                                                   int node)
 126 {
 127 #ifdef CONFIG_DEBUG_STACK_USAGE
 128         gfp_t mask = GFP_KERNEL | __GFP_ZERO;
 129 #else
 130         gfp_t mask = GFP_KERNEL;
 131 #endif
 132         struct page *page = alloc_pages_node(node, mask, THREAD_SIZE_ORDER);
 133
 134         return page ? page_address(page) : NULL;
 135 }
 136
 137 static inline void free_thread_info(struct thread_info *ti)
 138 {
 139         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
 140 }
 141 #endif
 142
 143 /* SLAB cache for signal_struct structures (tsk->signal) */
 144 static struct kmem_cache *signal_cachep;
 145
 146 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 147 struct kmem_cache *sighand_cachep;
 148
 149 /* SLAB cache for files_struct structures (tsk->files) */
 150 struct kmem_cache *files_cachep;
 151
 152 /* SLAB cache for fs_struct structures (tsk->fs) */
 153 struct kmem_cache *fs_cachep;
 154
 155 /* SLAB cache for vm_area_struct structures */
 156 struct kmem_cache *vm_area_cachep;
 157
 158 /* SLAB cache for mm_struct structures (tsk->mm) */
 159 static struct kmem_cache *mm_cachep;
 160
 161 static void account_kernel_stack(struct thread_info *ti, int account)
 162 {
 163         struct zone *zone = page_zone(virt_to_page(ti));
 164
 165         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
 166 }
 167
 168 void free_task(struct task_struct *tsk)
 169 {
 170         account_kernel_stack(tsk->stack, -1);
 171         free_thread_info(tsk->stack);
 172         rt_mutex_debug_task_free(tsk);
 173         ftrace_graph_exit_task(tsk);
 174         free_task_struct(tsk);
 175 }
 176 EXPORT_SYMBOL(free_task);
 177
 178 static inline void free_signal_struct(struct signal_struct *sig)
 179 {
 180         taskstats_tgid_free(sig);
 181         sched_autogroup_exit(sig);
 182         kmem_cache_free(signal_cachep, sig);
 183 }
 184
 185 static inline void put_signal_struct(struct signal_struct *sig)
 186 {
 187         if (atomic_dec_and_test(&sig->sigcnt))
 188                 free_signal_struct(sig);
 189 }
 190
 191 void __put_task_struct(struct task_struct *tsk)
 192 {
 193         WARN_ON(!tsk->exit_state);
 194         WARN_ON(atomic_read(&tsk->usage));
 195         WARN_ON(tsk == current);
 196
 197         security_task_free(tsk);
 198         exit_creds(tsk);
 199         delayacct_tsk_free(tsk);
 200         put_signal_struct(tsk->signal);
 201
 202         if (!profile_handoff_task(tsk))
 203                 free_task(tsk);
 204 }
 205 EXPORT_SYMBOL_GPL(__put_task_struct);
 206
 207 /*
 208  * macro override instead of weak attribute alias, to workaround
 209  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
 210  */
 211 #ifndef arch_task_cache_init
 212 #define arch_task_cache_init()
 213 #endif
 214
 215 void __init fork_init(unsigned long mempages)
 216 {
 217 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 218 #ifndef ARCH_MIN_TASKALIGN
 219 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 220 #endif
 221         /* create a slab on which task_structs can be allocated */
 222         task_struct_cachep =
 223                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 224                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
 225 #endif
 226
 227         /* do the arch specific task caches init */
 228         arch_task_cache_init();
 229
 230         /*
 231          * The default maximum number of threads is set to a safe
 232          * value: the thread structures can take up at most half
 233          * of memory.
 234          */
 235         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 236
 237         /*
 238          * we need to allow at least 20 threads to boot a system
 239          */
 240         if (max_threads < 20)
 241                 max_threads = 20;
 242
 243         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 244         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 245         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 246                 init_task.signal->rlim[RLIMIT_NPROC];
 247 }
 248
 249 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
 250                                                struct task_struct *src)
 251 {
 252         *dst = *src;
 253         return 0;
 254 }
 255
 256 static struct task_struct *dup_task_struct(struct task_struct *orig)
 257 {
 258         struct task_struct *tsk;
 259         struct thread_info *ti;
 260         unsigned long *stackend;
 261         int node = tsk_fork_get_node(orig);
 262         int err;
 263
 264         prepare_to_copy(orig);
 265
 266         tsk = alloc_task_struct_node(node);
 267         if (!tsk)
 268                 return NULL;
 269
 270         ti = alloc_thread_info_node(tsk, node);
 271         if (!ti) {
 272                 free_task_struct(tsk);
 273                 return NULL;
 274         }
 275
 276         err = arch_dup_task_struct(tsk, orig);
 277         if (err)
 278                 goto out;
 279
 280         tsk->stack = ti;
 281
 282         setup_thread_stack(tsk, orig);
 283         clear_user_return_notifier(tsk);
 284         clear_tsk_need_resched(tsk);
 285         stackend = end_of_stack(tsk);
 286         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 287
 288 #ifdef CONFIG_CC_STACKPROTECTOR
 289         tsk->stack_canary = get_random_int();
 290 #endif
 291
 292         /*
 293          * One for us, one for whoever does the "release_task()" (usually
 294          * parent)
 295          */
 296         atomic_set(&tsk->usage, 2);
 297 #ifdef CONFIG_BLK_DEV_IO_TRACE
 298         tsk->btrace_seq = 0;
 299 #endif
 300         tsk->splice_pipe = NULL;
 301
 302         account_kernel_stack(ti, 1);
 303
 304         return tsk;
 305
 306 out:
 307         free_thread_info(ti);
 308         free_task_struct(tsk);
 309         return NULL;
 310 }
 311
 312 #ifdef CONFIG_MMU
 313 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 314 {
 315         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 316         struct rb_node **rb_link, *rb_parent;
 317         int retval;
 318         unsigned long charge;
 319         struct mempolicy *pol;
 320
 321         down_write(&oldmm->mmap_sem);
 322         flush_cache_dup_mm(oldmm);
 323         /*
 324          * Not linked in yet - no deadlock potential:
 325          */
 326         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 327
 328         mm->locked_vm = 0;
 329         mm->mmap = NULL;
 330         mm->mmap_cache = NULL;
 331         mm->free_area_cache = oldmm->mmap_base;
 332         mm->cached_hole_size = ~0UL;
 333         mm->map_count = 0;
 334         cpumask_clear(mm_cpumask(mm));
 335         mm->mm_rb = RB_ROOT;
 336         rb_link = &mm->mm_rb.rb_node;
 337         rb_parent = NULL;
 338         pprev = &mm->mmap;
 339         retval = ksm_fork(mm, oldmm);
 340         if (retval)
 341                 goto out;
 342         retval = khugepaged_fork(mm, oldmm);
 343         if (retval)
 344                 goto out;
 345
 346         prev = NULL;
 347         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 348                 struct file *file;
 349
 350                 if (mpnt->vm_flags & VM_DONTCOPY) {
 351                         long pages = vma_pages(mpnt);
 352                         mm->total_vm -= pages;
 353                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 354                                                                 -pages);
 355                         continue;
 356                 }
 357                 charge = 0;
 358                 if (mpnt->vm_flags & VM_ACCOUNT) {
 359                         unsigned long len;
 360                         len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 361                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 362                                 goto fail_nomem;
 363                         charge = len;
 364                 }
 365                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 366                 if (!tmp)
 367                         goto fail_nomem;
 368                 *tmp = *mpnt;
 369                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 370                 pol = mpol_dup(vma_policy(mpnt));
 371                 retval = PTR_ERR(pol);
 372                 if (IS_ERR(pol))
 373                         goto fail_nomem_policy;
 374                 vma_set_policy(tmp, pol);
 375                 tmp->vm_mm = mm;
 376                 if (anon_vma_fork(tmp, mpnt))
 377                         goto fail_nomem_anon_vma_fork;
 378                 tmp->vm_flags &= ~VM_LOCKED;
 379                 tmp->vm_next = tmp->vm_prev = NULL;
 380                 file = tmp->vm_file;
 381                 if (file) {
 382                         struct inode *inode = file->f_path.dentry->d_inode;
 383                         struct address_space *mapping = file->f_mapping;
 384
 385                         get_file(file);
 386                         if (tmp->vm_flags & VM_DENYWRITE)
 387                                 atomic_dec(&inode->i_writecount);
 388                         mutex_lock(&mapping->i_mmap_mutex);
 389                         if (tmp->vm_flags & VM_SHARED)
 390                                 mapping->i_mmap_writable++;
 391                         flush_dcache_mmap_lock(mapping);
 392                         /* insert tmp into the share list, just after mpnt */
 393                         vma_prio_tree_add(tmp, mpnt);
 394                         flush_dcache_mmap_unlock(mapping);
 395                         mutex_unlock(&mapping->i_mmap_mutex);
 396                 }
 397
 398                 /*
 399                  * Clear hugetlb-related page reserves for children. This only
 400                  * affects MAP_PRIVATE mappings. Faults generated by the child
 401                  * are not guaranteed to succeed, even if read-only
 402                  */
 403                 if (is_vm_hugetlb_page(tmp))
 404                         reset_vma_resv_huge_pages(tmp);
 405
 406                 /*
 407                  * Link in the new vma and copy the page table entries.
 408                  */
 409                 *pprev = tmp;
 410                 pprev = &tmp->vm_next;
 411                 tmp->vm_prev = prev;
 412                 prev = tmp;
 413
 414                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 415                 rb_link = &tmp->vm_rb.rb_right;
 416                 rb_parent = &tmp->vm_rb;
 417
 418                 mm->map_count++;
 419                 retval = copy_page_range(mm, oldmm, mpnt);
 420
 421                 if (tmp->vm_ops && tmp->vm_ops->open)
 422                         tmp->vm_ops->open(tmp);
 423
 424                 if (retval)
 425                         goto out;
 426         }
 427         /* a new mm has just been created */
 428         arch_dup_mmap(oldmm, mm);
 429         retval = 0;
 430 out:
 431         up_write(&mm->mmap_sem);
 432         flush_tlb_mm(oldmm);
 433         up_write(&oldmm->mmap_sem);
 434         return retval;
 435 fail_nomem_anon_vma_fork:
 436         mpol_put(pol);
 437 fail_nomem_policy:
 438         kmem_cache_free(vm_area_cachep, tmp);
 439 fail_nomem:
 440         retval = -ENOMEM;
 441         vm_unacct_memory(charge);
 442         goto out;
 443 }
 444
 445 static inline int mm_alloc_pgd(struct mm_struct *mm)
 446 {
 447         mm->pgd = pgd_alloc(mm);
 448         if (unlikely(!mm->pgd))
 449                 return -ENOMEM;
 450         return 0;
 451 }
 452
 453 static inline void mm_free_pgd(struct mm_struct *mm)
 454 {
 455         pgd_free(mm, mm->pgd);
 456 }
 457 #else
 458 #define dup_mmap(mm, oldmm)     (0)
 459 #define mm_alloc_pgd(mm)        (0)
 460 #define mm_free_pgd(mm)
 461 #endif /* CONFIG_MMU */
 462
 463 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 464
 465 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 466 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 467
 468 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 469
 470 static int __init coredump_filter_setup(char *s)
 471 {
 472         default_dump_filter =
 473                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 474                 MMF_DUMP_FILTER_MASK;
 475         return 1;
 476 }
 477
 478 __setup("coredump_filter=", coredump_filter_setup);
 479
 480 #include <linux/init_task.h>
 481
 482 static void mm_init_aio(struct mm_struct *mm)
 483 {
 484 #ifdef CONFIG_AIO
 485         spin_lock_init(&mm->ioctx_lock);
 486         INIT_HLIST_HEAD(&mm->ioctx_list);
 487 #endif
 488 }
 489
 490 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
 491 {
 492         atomic_set(&mm->mm_users, 1);
 493         atomic_set(&mm->mm_count, 1);
 494         init_rwsem(&mm->mmap_sem);
 495         INIT_LIST_HEAD(&mm->mmlist);
 496         mm->flags = (current->mm) ?
 497                 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
 498         mm->core_state = NULL;
 499         mm->nr_ptes = 0;
 500         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 501         spin_lock_init(&mm->page_table_lock);
 502         mm->free_area_cache = TASK_UNMAPPED_BASE;
 503         mm->cached_hole_size = ~0UL;
 504         mm_init_aio(mm);
 505         mm_init_owner(mm, p);
 506
 507         if (likely(!mm_alloc_pgd(mm))) {
 508                 mm->def_flags = 0;
 509                 mmu_notifier_mm_init(mm);
 510                 return mm;
 511         }
 512
 513         free_mm(mm);
 514         return NULL;
 515 }
 516
 517 static void check_mm(struct mm_struct *mm)
 518 {
 519         int i;
 520
 521         for (i = 0; i < NR_MM_COUNTERS; i++) {
 522                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 523
 524                 if (unlikely(x))
 525                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 526                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 527         }
 528
 529 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 530         VM_BUG_ON(mm->pmd_huge_pte);
 531 #endif
 532 }
 533
 534 /*
 535  * Allocate and initialize an mm_struct.
 536  */
 537 struct mm_struct *mm_alloc(void)
 538 {
 539         struct mm_struct *mm;
 540
 541         mm = allocate_mm();
 542         if (!mm)
 543                 return NULL;
 544
 545         memset(mm, 0, sizeof(*mm));
 546         mm_init_cpumask(mm);
 547         return mm_init(mm, current);
 548 }
 549
 550 /*
 551  * Called when the last reference to the mm
 552  * is dropped: either by a lazy thread or by
 553  * mmput. Free the page directory and the mm.
 554  */
 555 void __mmdrop(struct mm_struct *mm)
 556 {
 557         BUG_ON(mm == &init_mm);
 558         mm_free_pgd(mm);
 559         destroy_context(mm);
 560         mmu_notifier_mm_destroy(mm);
 561         check_mm(mm);
 562         free_mm(mm);
 563 }
 564 EXPORT_SYMBOL_GPL(__mmdrop);
 565
 566 /*
 567  * Decrement the use count and release all resources for an mm.
 568  */
 569 void mmput(struct mm_struct *mm)
 570 {
 571         might_sleep();
 572
 573         if (atomic_dec_and_test(&mm->mm_users)) {
 574                 exit_aio(mm);
 575                 ksm_exit(mm);
 576                 khugepaged_exit(mm); /* must run before exit_mmap */
 577                 exit_mmap(mm);
 578                 set_mm_exe_file(mm, NULL);
 579                 if (!list_empty(&mm->mmlist)) {
 580                         spin_lock(&mmlist_lock);
 581                         list_del(&mm->mmlist);
 582                         spin_unlock(&mmlist_lock);
 583                 }
 584                 put_swap_token(mm);
 585                 if (mm->binfmt)
 586                         module_put(mm->binfmt->module);
 587                 mmdrop(mm);
 588         }
 589 }
 590 EXPORT_SYMBOL_GPL(mmput);
 591
 592 /*
 593  * We added or removed a vma mapping the executable. The vmas are only mapped
 594  * during exec and are not mapped with the mmap system call.
 595  * Callers must hold down_write() on the mm's mmap_sem for these
 596  */
 597 void added_exe_file_vma(struct mm_struct *mm)
 598 {
 599         mm->num_exe_file_vmas++;
 600 }
 601
 602 void removed_exe_file_vma(struct mm_struct *mm)
 603 {
 604         mm->num_exe_file_vmas--;
 605         if ((mm->num_exe_file_vmas == 0) && mm->exe_file) {
 606                 fput(mm->exe_file);
 607                 mm->exe_file = NULL;
 608         }
 609
 610 }
 611
 612 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 613 {
 614         if (new_exe_file)
 615                 get_file(new_exe_file);
 616         if (mm->exe_file)
 617                 fput(mm->exe_file);
 618         mm->exe_file = new_exe_file;
 619         mm->num_exe_file_vmas = 0;
 620 }
 621
 622 struct file *get_mm_exe_file(struct mm_struct *mm)
 623 {
 624         struct file *exe_file;
 625
 626         /* We need mmap_sem to protect against races with removal of
 627          * VM_EXECUTABLE vmas */
 628         down_read(&mm->mmap_sem);
 629         exe_file = mm->exe_file;
 630         if (exe_file)
 631                 get_file(exe_file);
 632         up_read(&mm->mmap_sem);
 633         return exe_file;
 634 }
 635
 636 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
 637 {
 638         /* It's safe to write the exe_file pointer without exe_file_lock because
 639          * this is called during fork when the task is not yet in /proc */
 640         newmm->exe_file = get_mm_exe_file(oldmm);
 641 }
 642
 643 /**
 644  * get_task_mm - acquire a reference to the task's mm
 645  *
 646  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 647  * this kernel workthread has transiently adopted a user mm with use_mm,
 648  * to do its AIO) is not set and if so returns a reference to it, after
 649  * bumping up the use count.  User must release the mm via mmput()
 650  * after use.  Typically used by /proc and ptrace.
 651  */
 652 struct mm_struct *get_task_mm(struct task_struct *task)
 653 {
 654         struct mm_struct *mm;
 655
 656         task_lock(task);
 657         mm = task->mm;
 658         if (mm) {
 659                 if (task->flags & PF_KTHREAD)
 660                         mm = NULL;
 661                 else
 662                         atomic_inc(&mm->mm_users);
 663         }
 664         task_unlock(task);
 665         return mm;
 666 }
 667 EXPORT_SYMBOL_GPL(get_task_mm);
 668
 669 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
 670 {
 671         struct mm_struct *mm;
 672         int err;
 673
 674         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
 675         if (err)
 676                 return ERR_PTR(err);
 677
 678         mm = get_task_mm(task);
 679         if (mm && mm != current->mm &&
 680                         !ptrace_may_access(task, mode)) {
 681                 mmput(mm);
 682                 mm = ERR_PTR(-EACCES);
 683         }
 684         mutex_unlock(&task->signal->cred_guard_mutex);
 685
 686         return mm;
 687 }
 688
 689 static void complete_vfork_done(struct task_struct *tsk)
 690 {
 691         struct completion *vfork;
 692
 693         task_lock(tsk);
 694         vfork = tsk->vfork_done;
 695         if (likely(vfork)) {
 696                 tsk->vfork_done = NULL;
 697                 complete(vfork);
 698         }
 699         task_unlock(tsk);
 700 }
 701
 702 static int wait_for_vfork_done(struct task_struct *child,
 703                                 struct completion *vfork)
 704 {
 705         int killed;
 706
 707         freezer_do_not_count();
 708         killed = wait_for_completion_killable(vfork);
 709         freezer_count();
 710
 711         if (killed) {
 712                 task_lock(child);
 713                 child->vfork_done = NULL;
 714                 task_unlock(child);
 715         }
 716
 717         put_task_struct(child);
 718         return killed;
 719 }
 720
 721 /* Please note the differences between mmput and mm_release.
 722  * mmput is called whenever we stop holding onto a mm_struct,
 723  * error success whatever.
 724  *
 725  * mm_release is called after a mm_struct has been removed
 726  * from the current process.
 727  *
 728  * This difference is important for error handling, when we
 729  * only half set up a mm_struct for a new process and need to restore
 730  * the old one.  Because we mmput the new mm_struct before
 731  * restoring the old one. . .
 732  * Eric Biederman 10 January 1998
 733  */
 734 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 735 {
 736         /* Get rid of any futexes when releasing the mm */
 737 #ifdef CONFIG_FUTEX
 738         if (unlikely(tsk->robust_list)) {
 739                 exit_robust_list(tsk);
 740                 tsk->robust_list = NULL;
 741         }
 742 #ifdef CONFIG_COMPAT
 743         if (unlikely(tsk->compat_robust_list)) {
 744                 compat_exit_robust_list(tsk);
 745                 tsk->compat_robust_list = NULL;
 746         }
 747 #endif
 748         if (unlikely(!list_empty(&tsk->pi_state_list)))
 749                 exit_pi_state_list(tsk);
 750 #endif
 751
 752         /* Get rid of any cached register state */
 753         deactivate_mm(tsk, mm);
 754
 755         if (tsk->vfork_done)
 756                 complete_vfork_done(tsk);
 757
 758         /*
 759          * If we're exiting normally, clear a user-space tid field if
 760          * requested.  We leave this alone when dying by signal, to leave
 761          * the value intact in a core dump, and to save the unnecessary
 762          * trouble, say, a killed vfork parent shouldn't touch this mm.
 763          * Userland only wants this done for a sys_exit.
 764          */
 765         if (tsk->clear_child_tid) {
 766                 if (!(tsk->flags & PF_SIGNALED) &&
 767                     atomic_read(&mm->mm_users) > 1) {
 768                         /*
 769                          * We don't check the error code - if userspace has
 770                          * not set up a proper pointer then tough luck.
 771                          */
 772                         put_user(0, tsk->clear_child_tid);
 773                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 774                                         1, NULL, NULL, 0);
 775                 }
 776                 tsk->clear_child_tid = NULL;
 777         }
 778 }
 779
 780 /*
 781  * Allocate a new mm structure and copy contents from the
 782  * mm structure of the passed in task structure.
 783  */
 784 struct mm_struct *dup_mm(struct task_struct *tsk)
 785 {
 786         struct mm_struct *mm, *oldmm = current->mm;
 787         int err;
 788
 789         if (!oldmm)
 790                 return NULL;
 791
 792         mm = allocate_mm();
 793         if (!mm)
 794                 goto fail_nomem;
 795
 796         memcpy(mm, oldmm, sizeof(*mm));
 797         mm_init_cpumask(mm);
 798
 799         /* Initializing for Swap token stuff */
 800         mm->token_priority = 0;
 801         mm->last_interval = 0;
 802
 803 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 804         mm->pmd_huge_pte = NULL;
 805 #endif
 806
 807         if (!mm_init(mm, tsk))
 808                 goto fail_nomem;
 809
 810         if (init_new_context(tsk, mm))
 811                 goto fail_nocontext;
 812
 813         dup_mm_exe_file(oldmm, mm);
 814
 815         err = dup_mmap(mm, oldmm);
 816         if (err)
 817                 goto free_pt;
 818
 819         mm->hiwater_rss = get_mm_rss(mm);
 820         mm->hiwater_vm = mm->total_vm;
 821
 822         if (mm->binfmt && !try_module_get(mm->binfmt->module))
 823                 goto free_pt;
 824
 825         return mm;
 826
 827 free_pt:
 828         /* don't put binfmt in mmput, we haven't got module yet */
 829         mm->binfmt = NULL;
 830         mmput(mm);
 831
 832 fail_nomem:
 833         return NULL;
 834
 835 fail_nocontext:
 836         /*
 837          * If init_new_context() failed, we cannot use mmput() to free the mm
 838          * because it calls destroy_context()
 839          */
 840         mm_free_pgd(mm);
 841         free_mm(mm);
 842         return NULL;
 843 }
 844
 845 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
 846 {
 847         struct mm_struct *mm, *oldmm;
 848         int retval;
 849
 850         tsk->min_flt = tsk->maj_flt = 0;
 851         tsk->nvcsw = tsk->nivcsw = 0;
 852 #ifdef CONFIG_DETECT_HUNG_TASK
 853         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
 854 #endif
 855
 856         tsk->mm = NULL;
 857         tsk->active_mm = NULL;
 858
 859         /*
 860          * Are we cloning a kernel thread?
 861          *
 862          * We need to steal a active VM for that..
 863          */
 864         oldmm = current->mm;
 865         if (!oldmm)
 866                 return 0;
 867
 868         if (clone_flags & CLONE_VM) {
 869                 atomic_inc(&oldmm->mm_users);
 870                 mm = oldmm;
 871                 goto good_mm;
 872         }
 873
 874         retval = -ENOMEM;
 875         mm = dup_mm(tsk);
 876         if (!mm)
 877                 goto fail_nomem;
 878
 879 good_mm:
 880         /* Initializing for Swap token stuff */
 881         mm->token_priority = 0;
 882         mm->last_interval = 0;
 883
 884         tsk->mm = mm;
 885         tsk->active_mm = mm;
 886         return 0;
 887
 888 fail_nomem:
 889         return retval;
 890 }
 891
 892 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 893 {
 894         struct fs_struct *fs = current->fs;
 895         if (clone_flags & CLONE_FS) {
 896                 /* tsk->fs is already what we want */
 897                 spin_lock(&fs->lock);
 898                 if (fs->in_exec) {
 899                         spin_unlock(&fs->lock);
 900                         return -EAGAIN;
 901                 }
 902                 fs->users++;
 903                 spin_unlock(&fs->lock);
 904                 return 0;
 905         }
 906         tsk->fs = copy_fs_struct(fs);
 907         if (!tsk->fs)
 908                 return -ENOMEM;
 909         return 0;
 910 }
 911
 912 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
 913 {
 914         struct files_struct *oldf, *newf;
 915         int error = 0;
 916
 917         /*
 918          * A background process may not have any files ...
 919          */
 920         oldf = current->files;
 921         if (!oldf)
 922                 goto out;
 923
 924         if (clone_flags & CLONE_FILES) {
 925                 atomic_inc(&oldf->count);
 926                 goto out;
 927         }
 928
 929         newf = dup_fd(oldf, &error);
 930         if (!newf)
 931                 goto out;
 932
 933         tsk->files = newf;
 934         error = 0;
 935 out:
 936         return error;
 937 }
 938
 939 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
 940 {
 941 #ifdef CONFIG_BLOCK
 942         struct io_context *ioc = current->io_context;
 943         struct io_context *new_ioc;
 944
 945         if (!ioc)
 946                 return 0;
 947         /*
 948          * Share io context with parent, if CLONE_IO is set
 949          */
 950         if (clone_flags & CLONE_IO) {
 951                 tsk->io_context = ioc_task_link(ioc);
 952                 if (unlikely(!tsk->io_context))
 953                         return -ENOMEM;
 954         } else if (ioprio_valid(ioc->ioprio)) {
 955                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
 956                 if (unlikely(!new_ioc))
 957                         return -ENOMEM;
 958
 959                 new_ioc->ioprio = ioc->ioprio;
 960                 put_io_context(new_ioc);
 961         }
 962 #endif
 963         return 0;
 964 }
 965
 966 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
 967 {
 968         struct sighand_struct *sig;
 969
 970         if (clone_flags & CLONE_SIGHAND) {
 971                 atomic_inc(&current->sighand->count);
 972                 return 0;
 973         }
 974         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 975         rcu_assign_pointer(tsk->sighand, sig);
 976         if (!sig)
 977                 return -ENOMEM;
 978         atomic_set(&sig->count, 1);
 979         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 980         return 0;
 981 }
 982
 983 void __cleanup_sighand(struct sighand_struct *sighand)
 984 {
 985         if (atomic_dec_and_test(&sighand->count)) {
 986                 signalfd_cleanup(sighand);
 987                 kmem_cache_free(sighand_cachep, sighand);
 988         }
 989 }
 990
 991
 992 /*
 993  * Initialize POSIX timer handling for a thread group.
 994  */
 995 static void posix_cpu_timers_init_group(struct signal_struct *sig)
 996 {
 997         unsigned long cpu_limit;
 998
 999         /* Thread group counters. */
1000         thread_group_cputime_init(sig);
1001
1002         cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1003         if (cpu_limit != RLIM_INFINITY) {
1004                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1005                 sig->cputimer.running = 1;
1006         }
1007
1008         /* The timer lists. */
1009         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1010         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1011         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1012 }
1013
1014 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1015 {
1016         struct signal_struct *sig;
1017
1018         if (clone_flags & CLONE_THREAD)
1019                 return 0;
1020
1021         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1022         tsk->signal = sig;
1023         if (!sig)
1024                 return -ENOMEM;
1025
1026         sig->nr_threads = 1;
1027         atomic_set(&sig->live, 1);
1028         atomic_set(&sig->sigcnt, 1);
1029         init_waitqueue_head(&sig->wait_chldexit);
1030         if (clone_flags & CLONE_NEWPID)
1031                 sig->flags |= SIGNAL_UNKILLABLE;
1032         sig->curr_target = tsk;
1033         init_sigpending(&sig->shared_pending);
1034         INIT_LIST_HEAD(&sig->posix_timers);
1035
1036         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1037         sig->real_timer.function = it_real_fn;
1038
1039         task_lock(current->group_leader);
1040         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1041         task_unlock(current->group_leader);
1042
1043         posix_cpu_timers_init_group(sig);
1044
1045         tty_audit_fork(sig);
1046         sched_autogroup_fork(sig);
1047
1048 #ifdef CONFIG_CGROUPS
1049         init_rwsem(&sig->group_rwsem);
1050 #endif
1051
1052         sig->oom_adj = current->signal->oom_adj;
1053         sig->oom_score_adj = current->signal->oom_score_adj;
1054         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1055
1056         sig->has_child_subreaper = current->signal->has_child_subreaper ||
1057                                    current->signal->is_child_subreaper;
1058
1059         mutex_init(&sig->cred_guard_mutex);
1060
1061         return 0;
1062 }
1063
1064 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
1065 {
1066         unsigned long new_flags = p->flags;
1067
1068         new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1069         new_flags |= PF_FORKNOEXEC;
1070         p->flags = new_flags;
1071 }
1072
1073 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1074 {
1075         current->clear_child_tid = tidptr;
1076
1077         return task_pid_vnr(current);
1078 }
1079
1080 static void rt_mutex_init_task(struct task_struct *p)
1081 {
1082         raw_spin_lock_init(&p->pi_lock);
1083 #ifdef CONFIG_RT_MUTEXES
1084         plist_head_init(&p->pi_waiters);
1085         p->pi_blocked_on = NULL;
1086 #endif
1087 }
1088
1089 #ifdef CONFIG_MM_OWNER
1090 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1091 {
1092         mm->owner = p;
1093 }
1094 #endif /* CONFIG_MM_OWNER */
1095
1096 /*
1097  * Initialize POSIX timer handling for a single task.
1098  */
1099 static void posix_cpu_timers_init(struct task_struct *tsk)
1100 {
1101         tsk->cputime_expires.prof_exp = 0;
1102         tsk->cputime_expires.virt_exp = 0;
1103         tsk->cputime_expires.sched_exp = 0;
1104         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1105         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1106         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1107 }
1108
1109 /*
1110  * This creates a new process as a copy of the old one,
1111  * but does not actually start it yet.
1112  *
1113  * It copies the registers, and all the appropriate
1114  * parts of the process environment (as per the clone
1115  * flags). The actual kick-off is left to the caller.
1116  */
1117 static struct task_struct *copy_process(unsigned long clone_flags,
1118                                         unsigned long stack_start,
1119                                         struct pt_regs *regs,
1120                                         unsigned long stack_size,
1121                                         int __user *child_tidptr,
1122                                         struct pid *pid,
1123                                         int trace)
1124 {
1125         int retval;
1126         struct task_struct *p;
1127         int cgroup_callbacks_done = 0;
1128
1129         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1130                 return ERR_PTR(-EINVAL);
1131
1132         /*
1133          * Thread groups must share signals as well, and detached threads
1134          * can only be started up within the thread group.
1135          */
1136         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1137                 return ERR_PTR(-EINVAL);
1138
1139         /*
1140          * Shared signal handlers imply shared VM. By way of the above,
1141          * thread groups also imply shared VM. Blocking this case allows
1142          * for various simplifications in other code.
1143          */
1144         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1145                 return ERR_PTR(-EINVAL);
1146
1147         /*
1148          * Siblings of global init remain as zombies on exit since they are
1149          * not reaped by their parent (swapper). To solve this and to avoid
1150          * multi-rooted process trees, prevent global and container-inits
1151          * from creating siblings.
1152          */
1153         if ((clone_flags & CLONE_PARENT) &&
1154                                 current->signal->flags & SIGNAL_UNKILLABLE)
1155                 return ERR_PTR(-EINVAL);
1156
1157         retval = security_task_create(clone_flags);
1158         if (retval)
1159                 goto fork_out;
1160
1161         retval = -ENOMEM;
1162         p = dup_task_struct(current);
1163         if (!p)
1164                 goto fork_out;
1165
1166         ftrace_graph_init_task(p);
1167
1168         rt_mutex_init_task(p);
1169
1170 #ifdef CONFIG_PROVE_LOCKING
1171         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1172         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1173 #endif
1174         retval = -EAGAIN;
1175         if (atomic_read(&p->real_cred->user->processes) >=
1176                         task_rlimit(p, RLIMIT_NPROC)) {
1177                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1178                     p->real_cred->user != INIT_USER)
1179                         goto bad_fork_free;
1180         }
1181         current->flags &= ~PF_NPROC_EXCEEDED;
1182
1183         retval = copy_creds(p, clone_flags);
1184         if (retval < 0)
1185                 goto bad_fork_free;
1186
1187         /*
1188          * If multiple threads are within copy_process(), then this check
1189          * triggers too late. This doesn't hurt, the check is only there
1190          * to stop root fork bombs.
1191          */
1192         retval = -EAGAIN;
1193         if (nr_threads >= max_threads)
1194                 goto bad_fork_cleanup_count;
1195
1196         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1197                 goto bad_fork_cleanup_count;
1198
1199         p->did_exec = 0;
1200         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1201         copy_flags(clone_flags, p);
1202         INIT_LIST_HEAD(&p->children);
1203         INIT_LIST_HEAD(&p->sibling);
1204         rcu_copy_process(p);
1205         p->vfork_done = NULL;
1206         spin_lock_init(&p->alloc_lock);
1207
1208         init_sigpending(&p->pending);
1209
1210         p->utime = p->stime = p->gtime = 0;
1211         p->utimescaled = p->stimescaled = 0;
1212 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
1213         p->prev_utime = p->prev_stime = 0;
1214 #endif
1215 #if defined(SPLIT_RSS_COUNTING)
1216         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1217 #endif
1218
1219         p->default_timer_slack_ns = current->timer_slack_ns;
1220
1221         task_io_accounting_init(&p->ioac);
1222         acct_clear_integrals(p);
1223
1224         posix_cpu_timers_init(p);
1225
1226         do_posix_clock_monotonic_gettime(&p->start_time);
1227         p->real_start_time = p->start_time;
1228         monotonic_to_bootbased(&p->real_start_time);
1229         p->io_context = NULL;
1230         p->audit_context = NULL;
1231         if (clone_flags & CLONE_THREAD)
1232                 threadgroup_change_begin(current);
1233         cgroup_fork(p);
1234 #ifdef CONFIG_NUMA
1235         p->mempolicy = mpol_dup(p->mempolicy);
1236         if (IS_ERR(p->mempolicy)) {
1237                 retval = PTR_ERR(p->mempolicy);
1238                 p->mempolicy = NULL;
1239                 goto bad_fork_cleanup_cgroup;
1240         }
1241         mpol_fix_fork_child_flag(p);
1242 #endif
1243 #ifdef CONFIG_CPUSETS
1244         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1245         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1246         seqcount_init(&p->mems_allowed_seq);
1247 #endif
1248 #ifdef CONFIG_TRACE_IRQFLAGS
1249         p->irq_events = 0;
1250 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1251         p->hardirqs_enabled = 1;
1252 #else
1253         p->hardirqs_enabled = 0;
1254 #endif
1255         p->hardirq_enable_ip = 0;
1256         p->hardirq_enable_event = 0;
1257         p->hardirq_disable_ip = _THIS_IP_;
1258         p->hardirq_disable_event = 0;
1259         p->softirqs_enabled = 1;
1260         p->softirq_enable_ip = _THIS_IP_;
1261         p->softirq_enable_event = 0;
1262         p->softirq_disable_ip = 0;
1263         p->softirq_disable_event = 0;
1264         p->hardirq_context = 0;
1265         p->softirq_context = 0;
1266 #endif
1267 #ifdef CONFIG_LOCKDEP
1268         p->lockdep_depth = 0; /* no locks held yet */
1269         p->curr_chain_key = 0;
1270         p->lockdep_recursion = 0;
1271 #endif
1272
1273 #ifdef CONFIG_DEBUG_MUTEXES
1274         p->blocked_on = NULL; /* not blocked yet */
1275 #endif
1276 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
1277         p->memcg_batch.do_batch = 0;
1278         p->memcg_batch.memcg = NULL;
1279 #endif
1280
1281         /* Perform scheduler related setup. Assign this task to a CPU. */
1282         sched_fork(p);
1283
1284         retval = perf_event_init_task(p);
1285         if (retval)
1286                 goto bad_fork_cleanup_policy;
1287         retval = audit_alloc(p);
1288         if (retval)
1289                 goto bad_fork_cleanup_policy;
1290         /* copy all the process information */
1291         retval = copy_semundo(clone_flags, p);
1292         if (retval)
1293                 goto bad_fork_cleanup_audit;
1294         retval = copy_files(clone_flags, p);
1295         if (retval)
1296                 goto bad_fork_cleanup_semundo;
1297         retval = copy_fs(clone_flags, p);
1298         if (retval)
1299                 goto bad_fork_cleanup_files;
1300         retval = copy_sighand(clone_flags, p);
1301         if (retval)
1302                 goto bad_fork_cleanup_fs;
1303         retval = copy_signal(clone_flags, p);
1304         if (retval)
1305                 goto bad_fork_cleanup_sighand;
1306         retval = copy_mm(clone_flags, p);
1307         if (retval)
1308                 goto bad_fork_cleanup_signal;
1309         retval = copy_namespaces(clone_flags, p);
1310         if (retval)
1311                 goto bad_fork_cleanup_mm;
1312         retval = copy_io(clone_flags, p);
1313         if (retval)
1314                 goto bad_fork_cleanup_namespaces;
1315         retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1316         if (retval)
1317                 goto bad_fork_cleanup_io;
1318
1319         if (pid != &init_struct_pid) {
1320                 retval = -ENOMEM;
1321                 pid = alloc_pid(p->nsproxy->pid_ns);
1322                 if (!pid)
1323                         goto bad_fork_cleanup_io;
1324         }
1325
1326         p->pid = pid_nr(pid);
1327         p->tgid = p->pid;
1328         if (clone_flags & CLONE_THREAD)
1329                 p->tgid = current->tgid;
1330
1331         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1332         /*
1333          * Clear TID on mm_release()?
1334          */
1335         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1336 #ifdef CONFIG_BLOCK
1337         p->plug = NULL;
1338 #endif
1339 #ifdef CONFIG_FUTEX
1340         p->robust_list = NULL;
1341 #ifdef CONFIG_COMPAT
1342         p->compat_robust_list = NULL;
1343 #endif
1344         INIT_LIST_HEAD(&p->pi_state_list);
1345         p->pi_state_cache = NULL;
1346 #endif
1347         /*
1348          * sigaltstack should be cleared when sharing the same VM
1349          */
1350         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1351                 p->sas_ss_sp = p->sas_ss_size = 0;
1352
1353         /*
1354          * Syscall tracing and stepping should be turned off in the
1355          * child regardless of CLONE_PTRACE.
1356          */
1357         user_disable_single_step(p);
1358         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1359 #ifdef TIF_SYSCALL_EMU
1360         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1361 #endif
1362         clear_all_latency_tracing(p);
1363
1364         /* ok, now we should be set up.. */
1365         if (clone_flags & CLONE_THREAD)
1366                 p->exit_signal = -1;
1367         else if (clone_flags & CLONE_PARENT)
1368                 p->exit_signal = current->group_leader->exit_signal;
1369         else
1370                 p->exit_signal = (clone_flags & CSIGNAL);
1371
1372         p->pdeath_signal = 0;
1373         p->exit_state = 0;
1374
1375         p->nr_dirtied = 0;
1376         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1377         p->dirty_paused_when = 0;
1378
1379         /*
1380          * Ok, make it visible to the rest of the system.
1381          * We dont wake it up yet.
1382          */
1383         p->group_leader = p;
1384         INIT_LIST_HEAD(&p->thread_group);
1385
1386         /* Now that the task is set up, run cgroup callbacks if
1387          * necessary. We need to run them before the task is visible
1388          * on the tasklist. */
1389         cgroup_fork_callbacks(p);
1390         cgroup_callbacks_done = 1;
1391
1392         /* Need tasklist lock for parent etc handling! */
1393         write_lock_irq(&tasklist_lock);
1394
1395         /* CLONE_PARENT re-uses the old parent */
1396         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1397                 p->real_parent = current->real_parent;
1398                 p->parent_exec_id = current->parent_exec_id;
1399         } else {
1400                 p->real_parent = current;
1401                 p->parent_exec_id = current->self_exec_id;
1402         }
1403
1404         spin_lock(&current->sighand->siglock);
1405
1406         /*
1407          * Process group and session signals need to be delivered to just the
1408          * parent before the fork or both the parent and the child after the
1409          * fork. Restart if a signal comes in before we add the new process to
1410          * it's process group.
1411          * A fatal signal pending means that current will exit, so the new
1412          * thread can't slip out of an OOM kill (or normal SIGKILL).
1413         */
1414         recalc_sigpending();
1415         if (signal_pending(current)) {
1416                 spin_unlock(&current->sighand->siglock);
1417                 write_unlock_irq(&tasklist_lock);
1418                 retval = -ERESTARTNOINTR;
1419                 goto bad_fork_free_pid;
1420         }
1421
1422         if (clone_flags & CLONE_THREAD) {
1423                 current->signal->nr_threads++;
1424                 atomic_inc(&current->signal->live);
1425                 atomic_inc(&current->signal->sigcnt);
1426                 p->group_leader = current->group_leader;
1427                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1428         }
1429
1430         if (likely(p->pid)) {
1431                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1432
1433                 if (thread_group_leader(p)) {
1434                         if (is_child_reaper(pid))
1435                                 p->nsproxy->pid_ns->child_reaper = p;
1436
1437                         p->signal->leader_pid = pid;
1438                         p->signal->tty = tty_kref_get(current->signal->tty);
1439                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1440                         attach_pid(p, PIDTYPE_SID, task_session(current));
1441                         list_add_tail(&p->sibling, &p->real_parent->children);
1442                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1443                         __this_cpu_inc(process_counts);
1444                 }
1445                 attach_pid(p, PIDTYPE_PID, pid);
1446                 nr_threads++;
1447         }
1448
1449         total_forks++;
1450         spin_unlock(&current->sighand->siglock);
1451         write_unlock_irq(&tasklist_lock);
1452         proc_fork_connector(p);
1453         cgroup_post_fork(p);
1454         if (clone_flags & CLONE_THREAD)
1455                 threadgroup_change_end(current);
1456         perf_event_fork(p);
1457
1458         trace_task_newtask(p, clone_flags);
1459
1460         return p;
1461
1462 bad_fork_free_pid:
1463         if (pid != &init_struct_pid)
1464                 free_pid(pid);
1465 bad_fork_cleanup_io:
1466         if (p->io_context)
1467                 exit_io_context(p);
1468 bad_fork_cleanup_namespaces:
1469         if (unlikely(clone_flags & CLONE_NEWPID))
1470                 pid_ns_release_proc(p->nsproxy->pid_ns);
1471         exit_task_namespaces(p);
1472 bad_fork_cleanup_mm:
1473         if (p->mm)
1474                 mmput(p->mm);
1475 bad_fork_cleanup_signal:
1476         if (!(clone_flags & CLONE_THREAD))
1477                 free_signal_struct(p->signal);
1478 bad_fork_cleanup_sighand:
1479         __cleanup_sighand(p->sighand);
1480 bad_fork_cleanup_fs:
1481         exit_fs(p); /* blocking */
1482 bad_fork_cleanup_files:
1483         exit_files(p); /* blocking */
1484 bad_fork_cleanup_semundo:
1485         exit_sem(p);
1486 bad_fork_cleanup_audit:
1487         audit_free(p);
1488 bad_fork_cleanup_policy:
1489         perf_event_free_task(p);
1490 #ifdef CONFIG_NUMA
1491         mpol_put(p->mempolicy);
1492 bad_fork_cleanup_cgroup:
1493 #endif
1494         if (clone_flags & CLONE_THREAD)
1495                 threadgroup_change_end(current);
1496         cgroup_exit(p, cgroup_callbacks_done);
1497         delayacct_tsk_free(p);
1498         module_put(task_thread_info(p)->exec_domain->module);
1499 bad_fork_cleanup_count:
1500         atomic_dec(&p->cred->user->processes);
1501         exit_creds(p);
1502 bad_fork_free:
1503         free_task(p);
1504 fork_out:
1505         return ERR_PTR(retval);
1506 }
1507
1508 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1509 {
1510         memset(regs, 0, sizeof(struct pt_regs));
1511         return regs;
1512 }
1513
1514 static inline void init_idle_pids(struct pid_link *links)
1515 {
1516         enum pid_type type;
1517
1518         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1519                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1520                 links[type].pid = &init_struct_pid;
1521         }
1522 }
1523
1524 struct task_struct * __cpuinit fork_idle(int cpu)
1525 {
1526         struct task_struct *task;
1527         struct pt_regs regs;
1528
1529         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1530                             &init_struct_pid, 0);
1531         if (!IS_ERR(task)) {
1532                 init_idle_pids(task->pids);
1533                 init_idle(task, cpu);
1534         }
1535
1536         return task;
1537 }
1538
1539 /*
1540  *  Ok, this is the main fork-routine.
1541  *
1542  * It copies the process, and if successful kick-starts
1543  * it and waits for it to finish using the VM if required.
1544  */
1545 long do_fork(unsigned long clone_flags,
1546               unsigned long stack_start,
1547               struct pt_regs *regs,
1548               unsigned long stack_size,
1549               int __user *parent_tidptr,
1550               int __user *child_tidptr)
1551 {
1552         struct task_struct *p;
1553         int trace = 0;
1554         long nr;
1555
1556         /*
1557          * Do some preliminary argument and permissions checking before we
1558          * actually start allocating stuff
1559          */
1560         if (clone_flags & CLONE_NEWUSER) {
1561                 if (clone_flags & CLONE_THREAD)
1562                         return -EINVAL;
1563                 /* hopefully this check will go away when userns support is
1564                  * complete
1565                  */
1566                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1567                                 !capable(CAP_SETGID))
1568                         return -EPERM;
1569         }
1570
1571         /*
1572          * Determine whether and which event to report to ptracer.  When
1573          * called from kernel_thread or CLONE_UNTRACED is explicitly
1574          * requested, no event is reported; otherwise, report if the event
1575          * for the type of forking is enabled.
1576          */
1577         if (likely(user_mode(regs)) && !(clone_flags & CLONE_UNTRACED)) {
1578                 if (clone_flags & CLONE_VFORK)
1579                         trace = PTRACE_EVENT_VFORK;
1580                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1581                         trace = PTRACE_EVENT_CLONE;
1582                 else
1583                         trace = PTRACE_EVENT_FORK;
1584
1585                 if (likely(!ptrace_event_enabled(current, trace)))
1586                         trace = 0;
1587         }
1588
1589         p = copy_process(clone_flags, stack_start, regs, stack_size,
1590                          child_tidptr, NULL, trace);
1591         /*
1592          * Do this prior waking up the new thread - the thread pointer
1593          * might get invalid after that point, if the thread exits quickly.
1594          */
1595         if (!IS_ERR(p)) {
1596                 struct completion vfork;
1597
1598                 trace_sched_process_fork(current, p);
1599
1600                 nr = task_pid_vnr(p);
1601
1602                 if (clone_flags & CLONE_PARENT_SETTID)
1603                         put_user(nr, parent_tidptr);
1604
1605                 if (clone_flags & CLONE_VFORK) {
1606                         p->vfork_done = &vfork;
1607                         init_completion(&vfork);
1608                         get_task_struct(p);
1609                 }
1610
1611                 wake_up_new_task(p);
1612
1613                 /* forking complete and child started to run, tell ptracer */
1614                 if (unlikely(trace))
1615                         ptrace_event(trace, nr);
1616
1617                 if (clone_flags & CLONE_VFORK) {
1618                         if (!wait_for_vfork_done(p, &vfork))
1619                                 ptrace_event(PTRACE_EVENT_VFORK_DONE, nr);
1620                 }
1621         } else {
1622                 nr = PTR_ERR(p);
1623         }
1624         return nr;
1625 }
1626
1627 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1628 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1629 #endif
1630
1631 static void sighand_ctor(void *data)
1632 {
1633         struct sighand_struct *sighand = data;
1634
1635         spin_lock_init(&sighand->siglock);
1636         init_waitqueue_head(&sighand->signalfd_wqh);
1637 }
1638
1639 void __init proc_caches_init(void)
1640 {
1641         sighand_cachep = kmem_cache_create("sighand_cache",
1642                         sizeof(struct sighand_struct), 0,
1643                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1644                         SLAB_NOTRACK, sighand_ctor);
1645         signal_cachep = kmem_cache_create("signal_cache",
1646                         sizeof(struct signal_struct), 0,
1647                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1648         files_cachep = kmem_cache_create("files_cache",
1649                         sizeof(struct files_struct), 0,
1650                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1651         fs_cachep = kmem_cache_create("fs_cache",
1652                         sizeof(struct fs_struct), 0,
1653                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1654         /*
1655          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1656          * whole struct cpumask for the OFFSTACK case. We could change
1657          * this to *only* allocate as much of it as required by the
1658          * maximum number of CPU's we can ever have.  The cpumask_allocation
1659          * is at the end of the structure, exactly for that reason.
1660          */
1661         mm_cachep = kmem_cache_create("mm_struct",
1662                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1663                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1664         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1665         mmap_init();
1666         nsproxy_cache_init();
1667 }
1668
1669 /*
1670  * Check constraints on flags passed to the unshare system call.
1671  */
1672 static int check_unshare_flags(unsigned long unshare_flags)
1673 {
1674         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1675                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1676                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1677                 return -EINVAL;
1678         /*
1679          * Not implemented, but pretend it works if there is nothing to
1680          * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1681          * needs to unshare vm.
1682          */
1683         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1684                 /* FIXME: get_task_mm() increments ->mm_users */
1685                 if (atomic_read(&current->mm->mm_users) > 1)
1686                         return -EINVAL;
1687         }
1688
1689         return 0;
1690 }
1691
1692 /*
1693  * Unshare the filesystem structure if it is being shared
1694  */
1695 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1696 {
1697         struct fs_struct *fs = current->fs;
1698
1699         if (!(unshare_flags & CLONE_FS) || !fs)
1700                 return 0;
1701
1702         /* don't need lock here; in the worst case we'll do useless copy */
1703         if (fs->users == 1)
1704                 return 0;
1705
1706         *new_fsp = copy_fs_struct(fs);
1707         if (!*new_fsp)
1708                 return -ENOMEM;
1709
1710         return 0;
1711 }
1712
1713 /*
1714  * Unshare file descriptor table if it is being shared
1715  */
1716 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1717 {
1718         struct files_struct *fd = current->files;
1719         int error = 0;
1720
1721         if ((unshare_flags & CLONE_FILES) &&
1722             (fd && atomic_read(&fd->count) > 1)) {
1723                 *new_fdp = dup_fd(fd, &error);
1724                 if (!*new_fdp)
1725                         return error;
1726         }
1727
1728         return 0;
1729 }
1730
1731 /*
1732  * unshare allows a process to 'unshare' part of the process
1733  * context which was originally shared using clone.  copy_*
1734  * functions used by do_fork() cannot be used here directly
1735  * because they modify an inactive task_struct that is being
1736  * constructed. Here we are modifying the current, active,
1737  * task_struct.
1738  */
1739 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1740 {
1741         struct fs_struct *fs, *new_fs = NULL;
1742         struct files_struct *fd, *new_fd = NULL;
1743         struct nsproxy *new_nsproxy = NULL;
1744         int do_sysvsem = 0;
1745         int err;
1746
1747         err = check_unshare_flags(unshare_flags);
1748         if (err)
1749                 goto bad_unshare_out;
1750
1751         /*
1752          * If unsharing namespace, must also unshare filesystem information.
1753          */
1754         if (unshare_flags & CLONE_NEWNS)
1755                 unshare_flags |= CLONE_FS;
1756         /*
1757          * CLONE_NEWIPC must also detach from the undolist: after switching
1758          * to a new ipc namespace, the semaphore arrays from the old
1759          * namespace are unreachable.
1760          */
1761         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1762                 do_sysvsem = 1;
1763         err = unshare_fs(unshare_flags, &new_fs);
1764         if (err)
1765                 goto bad_unshare_out;
1766         err = unshare_fd(unshare_flags, &new_fd);
1767         if (err)
1768                 goto bad_unshare_cleanup_fs;
1769         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy, new_fs);
1770         if (err)
1771                 goto bad_unshare_cleanup_fd;
1772
1773         if (new_fs || new_fd || do_sysvsem || new_nsproxy) {
1774                 if (do_sysvsem) {
1775                         /*
1776                          * CLONE_SYSVSEM is equivalent to sys_exit().
1777                          */
1778                         exit_sem(current);
1779                 }
1780
1781                 if (new_nsproxy) {
1782                         switch_task_namespaces(current, new_nsproxy);
1783                         new_nsproxy = NULL;
1784                 }
1785
1786                 task_lock(current);
1787
1788                 if (new_fs) {
1789                         fs = current->fs;
1790                         spin_lock(&fs->lock);
1791                         current->fs = new_fs;
1792                         if (--fs->users)
1793                                 new_fs = NULL;
1794                         else
1795                                 new_fs = fs;
1796                         spin_unlock(&fs->lock);
1797                 }
1798
1799                 if (new_fd) {
1800                         fd = current->files;
1801                         current->files = new_fd;
1802                         new_fd = fd;
1803                 }
1804
1805                 task_unlock(current);
1806         }
1807
1808         if (new_nsproxy)
1809                 put_nsproxy(new_nsproxy);
1810
1811 bad_unshare_cleanup_fd:
1812         if (new_fd)
1813                 put_files_struct(new_fd);
1814
1815 bad_unshare_cleanup_fs:
1816         if (new_fs)
1817                 free_fs_struct(new_fs);
1818
1819 bad_unshare_out:
1820         return err;
1821 }
1822
1823 /*
1824  *      Helper to unshare the files of the current task.
1825  *      We don't want to expose copy_files internals to
1826  *      the exec layer of the kernel.
1827  */
1828
1829 int unshare_files(struct files_struct **displaced)
1830 {
1831         struct task_struct *task = current;
1832         struct files_struct *copy = NULL;
1833         int error;
1834
1835         error = unshare_fd(CLONE_FILES, &copy);
1836         if (error || !copy) {
1837                 *displaced = NULL;
1838                 return error;
1839         }
1840         *displaced = task->files;
1841         task_lock(task);
1842         task->files = copy;
1843         task_unlock(task);
1844         return 0;
1845 }